Elevenlabs Transcribe

Requirements:

• ELEVENLABS_API_KEY in the project's .env file
• uv installed (dependencies auto-install via PEP 723)

Before transcribing, verify:

1. uv is available (dependency installation is automatic via inline script metadata — no venv or manual pip install needed)

2. API key configured in the .env file where Claude is running:

   ELEVENLABS_API_KEY=your-key-here
   

3. Audio file exists and is a supported format (mp3, wav, mp4, m4a, ogg, flac, webm, etc.)

MUST stop if the API key is missing — inform the user to add it to their .env file.

Step 1: Parse user input

Extract the audio file path and any options from $ARGUMENTS or the user's message. Supported options:

• --output <path> or -o <path> — where to save the transcript
• --language <code> — ISO-639 language code (e.g., eng, spa, fra, deu, jpn, zho)
• --num-speakers <n> — max speakers in the audio (1-32)
• --keyterms "term1" "term2" — words/phrases to bias transcription towards
• --timestamps none|word|character — timestamp granularity
• --no-diarize — disable speaker identification
• --no-audio-events — disable audio event tagging
• --json — output full JSON response

Step 2: Validate the audio file

Confirm the file path exists. Expand ~ paths. The script handles validation automatically but check early for a clear error message.

Step 3: Check for API key

grep -q "ELEVENLABS_API_KEY=" .env 2>/dev/null && echo "API key configured" || echo "API key missing"

If missing, tell the user to add ELEVENLABS_API_KEY= to their .env file and stop.

Step 4: Run transcription

Dependencies are installed automatically by uv via inline script metadata (PEP 723). No venv or manual pip install needed.

Basic transcription (diarize + audio events + auto language):

uv run ~/.claude/skills/elevenlabs-transcribe/scripts/transcribe.py "<audio_file_path>"

With output file and options:

uv run ~/.claude/skills/elevenlabs-transcribe/scripts/transcribe.py "<audio_file_path>" --output transcript.txt --language eng --num-speakers 3

With key terms for better accuracy:

uv run ~/.claude/skills/elevenlabs-transcribe/scripts/transcribe.py "<audio_file_path>" --keyterms "technical term" "product name"

Full JSON response:

uv run ~/.claude/skills/elevenlabs-transcribe/scripts/transcribe.py "<audio_file_path>" --json --output result.json

Step 5: Present results

Format the transcription output cleanly for the user. If diarization is enabled, group text by speaker. Highlight any audio events detected. Example output:

[Speaker 0]: Hello, how are you doing today?
[Speaker 1]: I'm doing great, thanks for asking! (laughter)